NIFA: Nonlinear IMC for Efficient Transformer Inference on FPGAs
July 17, 2026
NIFA integrates nonlinear in-memory computing (IMC) into FPGA architectures to support dynamic matrix-matrix multiplication (DIMM). This overcomes the limitations of conventional IMC designs that only support static weights, significantly improving efficiency for Transformer-based models.
HOW THIS AFFECTS YOU
●
builderYou can achieve higher energy efficiency for Transformer inference on FPGA hardware.
●
researcherThis architecture addresses the hardware bottleneck of nonlinear operations in analog IMC designs.